Reinforcement Learning for Cyber Defense: Adaptive and Autonomous Security Systems

Abstract

The security threats have become so advanced and ubiquitous that traditional safety nets are often no longer enough to protect us from the fluid nature of these new forms of cyber assault. We present a state-of-the-art of reinforcement learning (RL) to create self-adaptable and autonomic security systems for cyber defense. Using RL, these systems can learn on-the-fly and update their defenses in real-time based on new threats. We survey current literature, propose an RL-based cyber defense framework and illustrate the applicability of these systems to real-world environments for widespread usage.